Background
The client is a leading credit card provider headquartered in the United States. Established over 150 years ago, the client has over 100 million active cards and currently features amongst the top 100 companies in the ‘Fortune 500’ list.
The client was facing issues of low collections from defaulting credit card accounts and was looking at partnering with an analytics expert for a reliable and effective solution. Post deliberations with several analytics firms, they reached out to Firstsource to improve collections from the top decile through better predictions of propensity-to-pay.
Business challenge
A very basic prioritization model was in use to identify accounts that had highest propensity-to-pay and a shift to a more robust system was therefore an urgency.
- Propensity-to-pay prediction in the top decile was just a little over 78%.
- The client was able to collect a meagre 2% of the total outstanding amount.
- The limited information available for analysis included – outstanding amount, residence state of card holder and ‘probability-to-pay’ score accessed by the client. Vital details necessary for a thorough analysis, such as credit history and demographics couldn’t be shared on account of confidentiality restrictions.
Firstsource solution
Analysis framework
The process, analysed a year of data and identified a number of segments within defined groups along with necessary variables for the model. It also fleshed out certain proxy variables such as number of days since last payment and CPS score among others. In terms of methodology, it performed clustering using the elbow method.
New prioritization model
After using logistic regression on all groups, the solution identified deciles for outstanding amount, number of card holders that had paid and the amount paid. The next phase then successfully created a prioritization model that could assign propensity to pay for each account leading to optimized sequencing.
